Gen AI Model Artist

Workman Labs; Greater London; Hybride; GBP 90 000 - 130 000; Plein temps

Résumé du poste

Workman Labs is seeking a talented engineer to train and fine-tune GenAI models that generate hyperrealistic media. You will curate datasets ensuring identity coverage and balance lighting and pose while maintaining data quality.

You will apply advanced fine-tuning strategies, evaluate results, and diagnose failures like identity drift or artifacts. Collaboration with R&D and robust experiment logging are essential.

Qualifications

• Experience with generative model training/fine-tuning (Stable Diffusion family or similar).
• Familiarity with UNIX/Linux environments and GPU training workflows.
• Critical eye for facial likeness and visual quality; able to judge and explain issues clearly.
• Excellent teamwork, communication, and problem-solving; calm under pressure.
• Nice to have: VFX or compositing experience (Nuke), Python + ML engineering, face-swap systems experience.

Responsabilités

• Train and fine-tune GenAI models to generate hyperrealistic media (image and/or video).
• Curate and prepare datasets for generative training: identity coverage, lighting/pose balance, captioning, filtering, de-duplication.
• Improve realism and likeness through fine-tuning strategy (adapters/LoRA-style approaches, full fine-tunes) and systematic evaluation.
• Diagnose generative failure modes: identity drift, artifacts, texture issues, uncanny results, overfitting, conditioning sensitivity.
• Manage multiple training runs with prompt/eval sets, side-by-side comparisons, regression checks, and experiment logs.
• Collaborate with R&D to innovate GenAI training techniques; monitor inference speed, VRAM, throughput, and consistency.

Connaissances

GenAI training

Linux/UNIX

Facial likeness evaluation

Teamwork & communication

Python ML engineering

Outils

Stable Diffusion

Nuke

LoRA fine-tuning

Python

Description du poste

Trains, fine-tunes, and deploys generative AI models at DNEG to create hyperrealistic synthetic media, focusing on identity preservation, controllability, realism, and temporal consistency for sequences.

Location

London, United Kingdom

Compared with 15 other Research roles: Common across similar roles Unique to this role

What you'll do

Responsibilities

• Train and fine-tune GenAI models to generate hyperrealistic synthetic media (image and/or video) Common
• Curate and prepare datasets for generative training: identity coverage, lighting/pose balance, captioning, filtering, de-duplication
• Improve realism and likeness through fine-tuning strategy (adapters/LoRA-style approaches, full fine-tunes) and systematic evaluation
• Diagnose generative failure modes: identity drift, artifacts, texture issues, uncanny results, overfitting, conditioning sensitivity Unique
• Manage multiple training runs with prompt/eval sets, side-by-side comparisons, regression checks, and experiment logs
• Collaborate with R&D to innovate GenAI training techniques; monitor inference speed, VRAM, throughput, and consistency

What you'll need

Experience

• Proven experience with generative model training/fine-tuning (Stable Diffusion family or similar diffusion/GAN/transformer models) Common
• Familiarity with UNIX/Linux environments and GPU training workflows
• Critical eye for facial likeness and visual quality; able to judge and explain issues clearly Unique
• Excellent teamwork, communication, and problem-solving; calm under pressure Unique
• Nice to have: VFX or compositing experience (Nuke), Python + ML engineering, face-swap systems experience Unique